2 * Kernel Probes (KProbes)
3 * arch/i386/kernel/kprobes.c
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19 * Copyright (C) IBM Corporation, 2002, 2004
21 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
22 * Probes initial implementation ( includes contributions from
24 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
25 * interface to access function arguments.
26 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
27 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
28 * <prasanna@in.ibm.com> added function-return probes.
31 #include <linux/config.h>
32 #include <linux/kprobes.h>
33 #include <linux/ptrace.h>
34 #include <linux/preempt.h>
35 #include <asm/cacheflush.h>
36 #include <asm/kdebug.h>
39 void jprobe_return_end(void);
41 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
42 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
45 * returns non-zero if opcode modifies the interrupt flag.
47 static inline int is_IF_modifier(kprobe_opcode_t opcode)
52 case 0xcf: /* iret/iretd */
53 case 0x9d: /* popf/popfd */
59 int __kprobes arch_prepare_kprobe(struct kprobe *p)
61 /* insn: must be on special executable page on i386. */
62 p->ainsn.insn = get_insn_slot();
66 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
71 void __kprobes arch_arm_kprobe(struct kprobe *p)
73 *p->addr = BREAKPOINT_INSTRUCTION;
74 flush_icache_range((unsigned long) p->addr,
75 (unsigned long) p->addr + sizeof(kprobe_opcode_t));
78 void __kprobes arch_disarm_kprobe(struct kprobe *p)
81 flush_icache_range((unsigned long) p->addr,
82 (unsigned long) p->addr + sizeof(kprobe_opcode_t));
85 void __kprobes arch_remove_kprobe(struct kprobe *p)
88 free_insn_slot(p->ainsn.insn);
92 static inline void save_previous_kprobe(struct kprobe_ctlblk *kcb)
94 kcb->prev_kprobe.kp = kprobe_running();
95 kcb->prev_kprobe.status = kcb->kprobe_status;
96 kcb->prev_kprobe.old_eflags = kcb->kprobe_old_eflags;
97 kcb->prev_kprobe.saved_eflags = kcb->kprobe_saved_eflags;
100 static inline void restore_previous_kprobe(struct kprobe_ctlblk *kcb)
102 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
103 kcb->kprobe_status = kcb->prev_kprobe.status;
104 kcb->kprobe_old_eflags = kcb->prev_kprobe.old_eflags;
105 kcb->kprobe_saved_eflags = kcb->prev_kprobe.saved_eflags;
108 static inline void set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
109 struct kprobe_ctlblk *kcb)
111 __get_cpu_var(current_kprobe) = p;
112 kcb->kprobe_saved_eflags = kcb->kprobe_old_eflags
113 = (regs->eflags & (TF_MASK | IF_MASK));
114 if (is_IF_modifier(p->opcode))
115 kcb->kprobe_saved_eflags &= ~IF_MASK;
118 static inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
120 regs->eflags |= TF_MASK;
121 regs->eflags &= ~IF_MASK;
122 /*single step inline if the instruction is an int3*/
123 if (p->opcode == BREAKPOINT_INSTRUCTION)
124 regs->eip = (unsigned long)p->addr;
126 regs->eip = (unsigned long)p->ainsn.insn;
129 /* Called with kretprobe_lock held */
130 void __kprobes arch_prepare_kretprobe(struct kretprobe *rp,
131 struct pt_regs *regs)
133 unsigned long *sara = (unsigned long *)®s->esp;
134 struct kretprobe_instance *ri;
136 if ((ri = get_free_rp_inst(rp)) != NULL) {
139 ri->ret_addr = (kprobe_opcode_t *) *sara;
141 /* Replace the return addr with trampoline addr */
142 *sara = (unsigned long) &kretprobe_trampoline;
151 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
152 * remain disabled thorough out this function.
154 static int __kprobes kprobe_handler(struct pt_regs *regs)
158 kprobe_opcode_t *addr = NULL;
160 struct kprobe_ctlblk *kcb;
163 * We don't want to be preempted for the entire
164 * duration of kprobe processing
167 kcb = get_kprobe_ctlblk();
169 /* Check if the application is using LDT entry for its code segment and
170 * calculate the address by reading the base address from the LDT entry.
172 if ((regs->xcs & 4) && (current->mm)) {
173 lp = (unsigned long *) ((unsigned long)((regs->xcs >> 3) * 8)
174 + (char *) current->mm->context.ldt);
175 addr = (kprobe_opcode_t *) (get_desc_base(lp) + regs->eip -
176 sizeof(kprobe_opcode_t));
178 addr = (kprobe_opcode_t *)(regs->eip - sizeof(kprobe_opcode_t));
180 /* Check we're not actually recursing */
181 if (kprobe_running()) {
182 p = get_kprobe(addr);
184 if (kcb->kprobe_status == KPROBE_HIT_SS &&
185 *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
186 regs->eflags &= ~TF_MASK;
187 regs->eflags |= kcb->kprobe_saved_eflags;
190 /* We have reentered the kprobe_handler(), since
191 * another probe was hit while within the handler.
192 * We here save the original kprobes variables and
193 * just single step on the instruction of the new probe
194 * without calling any user handlers.
196 save_previous_kprobe(kcb);
197 set_current_kprobe(p, regs, kcb);
198 kprobes_inc_nmissed_count(p);
199 prepare_singlestep(p, regs);
200 kcb->kprobe_status = KPROBE_REENTER;
203 if (regs->eflags & VM_MASK) {
204 /* We are in virtual-8086 mode. Return 0 */
207 if (*addr != BREAKPOINT_INSTRUCTION) {
208 /* The breakpoint instruction was removed by
209 * another cpu right after we hit, no further
210 * handling of this interrupt is appropriate
212 regs->eip -= sizeof(kprobe_opcode_t);
216 p = __get_cpu_var(current_kprobe);
217 if (p->break_handler && p->break_handler(p, regs)) {
224 p = get_kprobe(addr);
226 if (regs->eflags & VM_MASK) {
227 /* We are in virtual-8086 mode. Return 0 */
231 if (*addr != BREAKPOINT_INSTRUCTION) {
233 * The breakpoint instruction was removed right
234 * after we hit it. Another cpu has removed
235 * either a probepoint or a debugger breakpoint
236 * at this address. In either case, no further
237 * handling of this interrupt is appropriate.
238 * Back up over the (now missing) int3 and run
239 * the original instruction.
241 regs->eip -= sizeof(kprobe_opcode_t);
244 /* Not one of ours: let kernel handle it */
248 set_current_kprobe(p, regs, kcb);
249 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
251 if (p->pre_handler && p->pre_handler(p, regs))
252 /* handler has already set things up, so skip ss setup */
256 prepare_singlestep(p, regs);
257 kcb->kprobe_status = KPROBE_HIT_SS;
261 preempt_enable_no_resched();
266 * For function-return probes, init_kprobes() establishes a probepoint
267 * here. When a retprobed function returns, this probe is hit and
268 * trampoline_probe_handler() runs, calling the kretprobe's handler.
270 void kretprobe_trampoline_holder(void)
272 asm volatile ( ".global kretprobe_trampoline\n"
273 "kretprobe_trampoline: \n"
278 * Called when we hit the probe point at kretprobe_trampoline
280 int __kprobes trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
282 struct kretprobe_instance *ri = NULL;
283 struct hlist_head *head;
284 struct hlist_node *node, *tmp;
285 unsigned long flags, orig_ret_address = 0;
286 unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
288 spin_lock_irqsave(&kretprobe_lock, flags);
289 head = kretprobe_inst_table_head(current);
292 * It is possible to have multiple instances associated with a given
293 * task either because an multiple functions in the call path
294 * have a return probe installed on them, and/or more then one return
295 * return probe was registered for a target function.
297 * We can handle this because:
298 * - instances are always inserted at the head of the list
299 * - when multiple return probes are registered for the same
300 * function, the first instance's ret_addr will point to the
301 * real return address, and all the rest will point to
302 * kretprobe_trampoline
304 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
305 if (ri->task != current)
306 /* another task is sharing our hash bucket */
309 if (ri->rp && ri->rp->handler)
310 ri->rp->handler(ri, regs);
312 orig_ret_address = (unsigned long)ri->ret_addr;
315 if (orig_ret_address != trampoline_address)
317 * This is the real return address. Any other
318 * instances associated with this task are for
319 * other calls deeper on the call stack
324 BUG_ON(!orig_ret_address || (orig_ret_address == trampoline_address));
325 regs->eip = orig_ret_address;
327 reset_current_kprobe();
328 spin_unlock_irqrestore(&kretprobe_lock, flags);
329 preempt_enable_no_resched();
332 * By returning a non-zero value, we are telling
333 * kprobe_handler() that we don't want the post_handler
334 * to run (and have re-enabled preemption)
340 * Called after single-stepping. p->addr is the address of the
341 * instruction whose first byte has been replaced by the "int 3"
342 * instruction. To avoid the SMP problems that can occur when we
343 * temporarily put back the original opcode to single-step, we
344 * single-stepped a copy of the instruction. The address of this
345 * copy is p->ainsn.insn.
347 * This function prepares to return from the post-single-step
348 * interrupt. We have to fix up the stack as follows:
350 * 0) Except in the case of absolute or indirect jump or call instructions,
351 * the new eip is relative to the copied instruction. We need to make
352 * it relative to the original instruction.
354 * 1) If the single-stepped instruction was pushfl, then the TF and IF
355 * flags are set in the just-pushed eflags, and may need to be cleared.
357 * 2) If the single-stepped instruction was a call, the return address
358 * that is atop the stack is the address following the copied instruction.
359 * We need to make it the address following the original instruction.
361 static void __kprobes resume_execution(struct kprobe *p,
362 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
364 unsigned long *tos = (unsigned long *)®s->esp;
365 unsigned long next_eip = 0;
366 unsigned long copy_eip = (unsigned long)p->ainsn.insn;
367 unsigned long orig_eip = (unsigned long)p->addr;
369 switch (p->ainsn.insn[0]) {
370 case 0x9c: /* pushfl */
371 *tos &= ~(TF_MASK | IF_MASK);
372 *tos |= kcb->kprobe_old_eflags;
374 case 0xc3: /* ret/lret */
378 regs->eflags &= ~TF_MASK;
379 /* eip is already adjusted, no more changes required*/
381 case 0xe8: /* call relative - Fix return addr */
382 *tos = orig_eip + (*tos - copy_eip);
385 if ((p->ainsn.insn[1] & 0x30) == 0x10) {
386 /* call absolute, indirect */
387 /* Fix return addr; eip is correct. */
388 next_eip = regs->eip;
389 *tos = orig_eip + (*tos - copy_eip);
390 } else if (((p->ainsn.insn[1] & 0x31) == 0x20) || /* jmp near, absolute indirect */
391 ((p->ainsn.insn[1] & 0x31) == 0x21)) { /* jmp far, absolute indirect */
392 /* eip is correct. */
393 next_eip = regs->eip;
396 case 0xea: /* jmp absolute -- eip is correct */
397 next_eip = regs->eip;
403 regs->eflags &= ~TF_MASK;
405 regs->eip = next_eip;
407 regs->eip = orig_eip + (regs->eip - copy_eip);
412 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
413 * remain disabled thoroughout this function.
415 static inline int post_kprobe_handler(struct pt_regs *regs)
417 struct kprobe *cur = kprobe_running();
418 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
423 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
424 kcb->kprobe_status = KPROBE_HIT_SSDONE;
425 cur->post_handler(cur, regs, 0);
428 resume_execution(cur, regs, kcb);
429 regs->eflags |= kcb->kprobe_saved_eflags;
431 /*Restore back the original saved kprobes variables and continue. */
432 if (kcb->kprobe_status == KPROBE_REENTER) {
433 restore_previous_kprobe(kcb);
436 reset_current_kprobe();
438 preempt_enable_no_resched();
441 * if somebody else is singlestepping across a probe point, eflags
442 * will have TF set, in which case, continue the remaining processing
443 * of do_debug, as if this is not a probe hit.
445 if (regs->eflags & TF_MASK)
451 static inline int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
453 struct kprobe *cur = kprobe_running();
454 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
456 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
459 if (kcb->kprobe_status & KPROBE_HIT_SS) {
460 resume_execution(cur, regs, kcb);
461 regs->eflags |= kcb->kprobe_old_eflags;
463 reset_current_kprobe();
464 preempt_enable_no_resched();
470 * Wrapper routine to for handling exceptions.
472 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
473 unsigned long val, void *data)
475 struct die_args *args = (struct die_args *)data;
476 int ret = NOTIFY_DONE;
480 if (kprobe_handler(args->regs))
484 if (post_kprobe_handler(args->regs))
489 /* kprobe_running() needs smp_processor_id() */
491 if (kprobe_running() &&
492 kprobe_fault_handler(args->regs, args->trapnr))
502 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
504 struct jprobe *jp = container_of(p, struct jprobe, kp);
506 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
508 kcb->jprobe_saved_regs = *regs;
509 kcb->jprobe_saved_esp = ®s->esp;
510 addr = (unsigned long)(kcb->jprobe_saved_esp);
513 * TBD: As Linus pointed out, gcc assumes that the callee
514 * owns the argument space and could overwrite it, e.g.
515 * tailcall optimization. So, to be absolutely safe
516 * we also save and restore enough stack bytes to cover
519 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
520 MIN_STACK_SIZE(addr));
521 regs->eflags &= ~IF_MASK;
522 regs->eip = (unsigned long)(jp->entry);
526 void __kprobes jprobe_return(void)
528 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
530 asm volatile (" xchgl %%ebx,%%esp \n"
532 " .globl jprobe_return_end \n"
533 " jprobe_return_end: \n"
535 (kcb->jprobe_saved_esp):"memory");
538 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
540 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
541 u8 *addr = (u8 *) (regs->eip - 1);
542 unsigned long stack_addr = (unsigned long)(kcb->jprobe_saved_esp);
543 struct jprobe *jp = container_of(p, struct jprobe, kp);
545 if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
546 if (®s->esp != kcb->jprobe_saved_esp) {
547 struct pt_regs *saved_regs =
548 container_of(kcb->jprobe_saved_esp,
549 struct pt_regs, esp);
550 printk("current esp %p does not match saved esp %p\n",
551 ®s->esp, kcb->jprobe_saved_esp);
552 printk("Saved registers for jprobe %p\n", jp);
553 show_registers(saved_regs);
554 printk("Current registers\n");
555 show_registers(regs);
558 *regs = kcb->jprobe_saved_regs;
559 memcpy((kprobe_opcode_t *) stack_addr, kcb->jprobes_stack,
560 MIN_STACK_SIZE(stack_addr));
561 preempt_enable_no_resched();
567 static struct kprobe trampoline_p = {
568 .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
569 .pre_handler = trampoline_probe_handler
572 int __init arch_init_kprobes(void)
574 return register_kprobe(&trampoline_p);